2 * linux/arch/x86-64/kernel/setup.c
4 * Copyright (C) 1995 Linus Torvalds
6 * Nov 2001 Dave Jones <davej@suse.de>
7 * Forked from i386 setup code.
11 * This file handles the architecture-dependent parts of initialization
14 #include <linux/errno.h>
15 #include <linux/sched.h>
16 #include <linux/kernel.h>
18 #include <linux/stddef.h>
19 #include <linux/unistd.h>
20 #include <linux/ptrace.h>
21 #include <linux/slab.h>
22 #include <linux/user.h>
23 #include <linux/a.out.h>
24 #include <linux/tty.h>
25 #include <linux/ioport.h>
26 #include <linux/delay.h>
27 #include <linux/config.h>
28 #include <linux/init.h>
29 #include <linux/initrd.h>
30 #include <linux/highmem.h>
31 #include <linux/bootmem.h>
32 #include <linux/module.h>
33 #include <asm/processor.h>
34 #include <linux/console.h>
35 #include <linux/seq_file.h>
36 #include <linux/crash_dump.h>
37 #include <linux/root_dev.h>
38 #include <linux/pci.h>
39 #include <linux/acpi.h>
40 #include <linux/kallsyms.h>
41 #include <linux/edd.h>
42 #include <linux/mmzone.h>
43 #include <linux/kexec.h>
44 #include <linux/cpufreq.h>
45 #include <linux/dmi.h>
46 #include <linux/dma-mapping.h>
47 #include <linux/ctype.h>
50 #include <asm/uaccess.h>
51 #include <asm/system.h>
56 #include <video/edid.h>
59 #include <asm/mpspec.h>
60 #include <asm/mmu_context.h>
61 #include <asm/bootsetup.h>
62 #include <asm/proto.h>
63 #include <asm/setup.h>
64 #include <asm/mach_apic.h>
66 #include <asm/swiotlb.h>
67 #include <asm/sections.h>
74 struct cpuinfo_x86 boot_cpu_data __read_mostly;
76 unsigned long mmu_cr4_features;
79 EXPORT_SYMBOL(acpi_disabled);
81 extern int __initdata acpi_ht;
82 extern acpi_interrupt_flags acpi_sci_flags;
83 int __initdata acpi_force = 0;
86 int acpi_numa __initdata;
88 /* Boot loader ID as an integer, for the benefit of proc_dointvec */
91 unsigned long saved_video_mode;
97 char dmi_alloc_data[DMI_MAX_DATA];
102 struct screen_info screen_info;
103 struct sys_desc_table_struct {
104 unsigned short length;
105 unsigned char table[0];
108 struct edid_info edid_info;
111 extern int root_mountflags;
113 char command_line[COMMAND_LINE_SIZE];
115 struct resource standard_io_resources[] = {
116 { .name = "dma1", .start = 0x00, .end = 0x1f,
117 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
118 { .name = "pic1", .start = 0x20, .end = 0x21,
119 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
120 { .name = "timer0", .start = 0x40, .end = 0x43,
121 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
122 { .name = "timer1", .start = 0x50, .end = 0x53,
123 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
124 { .name = "keyboard", .start = 0x60, .end = 0x6f,
125 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
126 { .name = "dma page reg", .start = 0x80, .end = 0x8f,
127 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
128 { .name = "pic2", .start = 0xa0, .end = 0xa1,
129 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
130 { .name = "dma2", .start = 0xc0, .end = 0xdf,
131 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
132 { .name = "fpu", .start = 0xf0, .end = 0xff,
133 .flags = IORESOURCE_BUSY | IORESOURCE_IO }
136 #define STANDARD_IO_RESOURCES \
137 (sizeof standard_io_resources / sizeof standard_io_resources[0])
139 #define IORESOURCE_RAM (IORESOURCE_BUSY | IORESOURCE_MEM)
141 struct resource data_resource = {
142 .name = "Kernel data",
145 .flags = IORESOURCE_RAM,
147 struct resource code_resource = {
148 .name = "Kernel code",
151 .flags = IORESOURCE_RAM,
154 #define IORESOURCE_ROM (IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM)
156 static struct resource system_rom_resource = {
157 .name = "System ROM",
160 .flags = IORESOURCE_ROM,
163 static struct resource extension_rom_resource = {
164 .name = "Extension ROM",
167 .flags = IORESOURCE_ROM,
170 static struct resource adapter_rom_resources[] = {
171 { .name = "Adapter ROM", .start = 0xc8000, .end = 0,
172 .flags = IORESOURCE_ROM },
173 { .name = "Adapter ROM", .start = 0, .end = 0,
174 .flags = IORESOURCE_ROM },
175 { .name = "Adapter ROM", .start = 0, .end = 0,
176 .flags = IORESOURCE_ROM },
177 { .name = "Adapter ROM", .start = 0, .end = 0,
178 .flags = IORESOURCE_ROM },
179 { .name = "Adapter ROM", .start = 0, .end = 0,
180 .flags = IORESOURCE_ROM },
181 { .name = "Adapter ROM", .start = 0, .end = 0,
182 .flags = IORESOURCE_ROM }
185 #define ADAPTER_ROM_RESOURCES \
186 (sizeof adapter_rom_resources / sizeof adapter_rom_resources[0])
188 static struct resource video_rom_resource = {
192 .flags = IORESOURCE_ROM,
195 static struct resource video_ram_resource = {
196 .name = "Video RAM area",
199 .flags = IORESOURCE_RAM,
202 #define romsignature(x) (*(unsigned short *)(x) == 0xaa55)
204 static int __init romchecksum(unsigned char *rom, unsigned long length)
206 unsigned char *p, sum = 0;
208 for (p = rom; p < rom + length; p++)
213 static void __init probe_roms(void)
215 unsigned long start, length, upper;
220 upper = adapter_rom_resources[0].start;
221 for (start = video_rom_resource.start; start < upper; start += 2048) {
222 rom = isa_bus_to_virt(start);
223 if (!romsignature(rom))
226 video_rom_resource.start = start;
228 /* 0 < length <= 0x7f * 512, historically */
229 length = rom[2] * 512;
231 /* if checksum okay, trust length byte */
232 if (length && romchecksum(rom, length))
233 video_rom_resource.end = start + length - 1;
235 request_resource(&iomem_resource, &video_rom_resource);
239 start = (video_rom_resource.end + 1 + 2047) & ~2047UL;
244 request_resource(&iomem_resource, &system_rom_resource);
245 upper = system_rom_resource.start;
247 /* check for extension rom (ignore length byte!) */
248 rom = isa_bus_to_virt(extension_rom_resource.start);
249 if (romsignature(rom)) {
250 length = extension_rom_resource.end - extension_rom_resource.start + 1;
251 if (romchecksum(rom, length)) {
252 request_resource(&iomem_resource, &extension_rom_resource);
253 upper = extension_rom_resource.start;
257 /* check for adapter roms on 2k boundaries */
258 for (i = 0; i < ADAPTER_ROM_RESOURCES && start < upper; start += 2048) {
259 rom = isa_bus_to_virt(start);
260 if (!romsignature(rom))
263 /* 0 < length <= 0x7f * 512, historically */
264 length = rom[2] * 512;
266 /* but accept any length that fits if checksum okay */
267 if (!length || start + length > upper || !romchecksum(rom, length))
270 adapter_rom_resources[i].start = start;
271 adapter_rom_resources[i].end = start + length - 1;
272 request_resource(&iomem_resource, &adapter_rom_resources[i]);
274 start = adapter_rom_resources[i++].end & ~2047UL;
278 /* Check for full argument with no trailing characters */
279 static int fullarg(char *p, char *arg)
282 return !memcmp(p, arg, l) && (p[l] == 0 || isspace(p[l]));
285 static __init void parse_cmdline_early (char ** cmdline_p)
287 char c = ' ', *to = command_line, *from = COMMAND_LINE;
297 * If the BIOS enumerates physical processors before logical,
298 * maxcpus=N at enumeration-time can be used to disable HT.
300 else if (!memcmp(from, "maxcpus=", 8)) {
301 extern unsigned int maxcpus;
303 maxcpus = simple_strtoul(from + 8, NULL, 0);
307 /* "acpi=off" disables both ACPI table parsing and interpreter init */
308 if (fullarg(from,"acpi=off"))
311 if (fullarg(from, "acpi=force")) {
312 /* add later when we do DMI horrors: */
317 /* acpi=ht just means: do ACPI MADT parsing
318 at bootup, but don't enable the full ACPI interpreter */
319 if (fullarg(from, "acpi=ht")) {
324 else if (fullarg(from, "pci=noacpi"))
326 else if (fullarg(from, "acpi=noirq"))
329 else if (fullarg(from, "acpi_sci=edge"))
330 acpi_sci_flags.trigger = 1;
331 else if (fullarg(from, "acpi_sci=level"))
332 acpi_sci_flags.trigger = 3;
333 else if (fullarg(from, "acpi_sci=high"))
334 acpi_sci_flags.polarity = 1;
335 else if (fullarg(from, "acpi_sci=low"))
336 acpi_sci_flags.polarity = 3;
338 /* acpi=strict disables out-of-spec workarounds */
339 else if (fullarg(from, "acpi=strict")) {
342 #ifdef CONFIG_X86_IO_APIC
343 else if (fullarg(from, "acpi_skip_timer_override"))
344 acpi_skip_timer_override = 1;
348 if (fullarg(from, "disable_timer_pin_1"))
349 disable_timer_pin_1 = 1;
350 if (fullarg(from, "enable_timer_pin_1"))
351 disable_timer_pin_1 = -1;
353 if (fullarg(from, "nolapic") || fullarg(from, "disableapic")) {
354 clear_bit(X86_FEATURE_APIC, boot_cpu_data.x86_capability);
358 if (fullarg(from, "noapic"))
359 skip_ioapic_setup = 1;
361 if (fullarg(from,"apic")) {
362 skip_ioapic_setup = 0;
366 if (!memcmp(from, "mem=", 4))
367 parse_memopt(from+4, &from);
369 if (!memcmp(from, "memmap=", 7)) {
370 /* exactmap option is for used defined memory */
371 if (!memcmp(from+7, "exactmap", 8)) {
372 #ifdef CONFIG_CRASH_DUMP
373 /* If we are doing a crash dump, we
374 * still need to know the real mem
375 * size before original memory map is
378 saved_max_pfn = e820_end_of_ram();
386 parse_memmapopt(from+7, &from);
392 if (!memcmp(from, "numa=", 5))
396 if (!memcmp(from,"iommu=",6)) {
400 if (fullarg(from,"oops=panic"))
403 if (!memcmp(from, "noexec=", 7))
404 nonx_setup(from + 7);
407 /* crashkernel=size@addr specifies the location to reserve for
408 * a crash kernel. By reserving this memory we guarantee
409 * that linux never set's it up as a DMA target.
410 * Useful for holding code to do something appropriate
411 * after a kernel panic.
413 else if (!memcmp(from, "crashkernel=", 12)) {
414 unsigned long size, base;
415 size = memparse(from+12, &from);
417 base = memparse(from+1, &from);
418 /* FIXME: Do I want a sanity check
419 * to validate the memory range?
421 crashk_res.start = base;
422 crashk_res.end = base + size - 1;
427 #ifdef CONFIG_PROC_VMCORE
428 /* elfcorehdr= specifies the location of elf core header
429 * stored by the crashed kernel. This option will be passed
430 * by kexec loader to the capture kernel.
432 else if(!memcmp(from, "elfcorehdr=", 11))
433 elfcorehdr_addr = memparse(from+11, &from);
436 #ifdef CONFIG_HOTPLUG_CPU
437 else if (!memcmp(from, "additional_cpus=", 16))
438 setup_additional_cpus(from+16);
445 if (COMMAND_LINE_SIZE <= ++len)
450 printk(KERN_INFO "user-defined physical RAM map:\n");
451 e820_print_map("user");
454 *cmdline_p = command_line;
459 contig_initmem_init(unsigned long start_pfn, unsigned long end_pfn)
461 unsigned long bootmap_size, bootmap;
463 bootmap_size = bootmem_bootmap_pages(end_pfn)<<PAGE_SHIFT;
464 bootmap = find_e820_area(0, end_pfn<<PAGE_SHIFT, bootmap_size);
466 panic("Cannot find bootmem map of size %ld\n",bootmap_size);
467 bootmap_size = init_bootmem(bootmap >> PAGE_SHIFT, end_pfn);
468 e820_bootmem_free(NODE_DATA(0), 0, end_pfn << PAGE_SHIFT);
469 reserve_bootmem(bootmap, bootmap_size);
473 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
475 #ifdef CONFIG_EDD_MODULE
479 * copy_edd() - Copy the BIOS EDD information
480 * from boot_params into a safe place.
483 static inline void copy_edd(void)
485 memcpy(edd.mbr_signature, EDD_MBR_SIGNATURE, sizeof(edd.mbr_signature));
486 memcpy(edd.edd_info, EDD_BUF, sizeof(edd.edd_info));
487 edd.mbr_signature_nr = EDD_MBR_SIG_NR;
488 edd.edd_info_nr = EDD_NR;
491 static inline void copy_edd(void)
496 #define EBDA_ADDR_POINTER 0x40E
498 unsigned __initdata ebda_addr;
499 unsigned __initdata ebda_size;
501 static void discover_ebda(void)
504 * there is a real-mode segmented pointer pointing to the
505 * 4K EBDA area at 0x40E
507 ebda_addr = *(unsigned short *)EBDA_ADDR_POINTER;
510 ebda_size = *(unsigned short *)(unsigned long)ebda_addr;
512 /* Round EBDA up to pages */
516 ebda_size = round_up(ebda_size + (ebda_addr & ~PAGE_MASK), PAGE_SIZE);
517 if (ebda_size > 64*1024)
521 void __init setup_arch(char **cmdline_p)
523 unsigned long kernel_end;
525 ROOT_DEV = old_decode_dev(ORIG_ROOT_DEV);
526 screen_info = SCREEN_INFO;
527 edid_info = EDID_INFO;
528 saved_video_mode = SAVED_VIDEO_MODE;
529 bootloader_type = LOADER_TYPE;
531 #ifdef CONFIG_BLK_DEV_RAM
532 rd_image_start = RAMDISK_FLAGS & RAMDISK_IMAGE_START_MASK;
533 rd_prompt = ((RAMDISK_FLAGS & RAMDISK_PROMPT_FLAG) != 0);
534 rd_doload = ((RAMDISK_FLAGS & RAMDISK_LOAD_FLAG) != 0);
536 setup_memory_region();
539 if (!MOUNT_ROOT_RDONLY)
540 root_mountflags &= ~MS_RDONLY;
541 init_mm.start_code = (unsigned long) &_text;
542 init_mm.end_code = (unsigned long) &_etext;
543 init_mm.end_data = (unsigned long) &_edata;
544 init_mm.brk = (unsigned long) &_end;
546 code_resource.start = virt_to_phys(&_text);
547 code_resource.end = virt_to_phys(&_etext)-1;
548 data_resource.start = virt_to_phys(&_etext);
549 data_resource.end = virt_to_phys(&_edata)-1;
551 parse_cmdline_early(cmdline_p);
553 early_identify_cpu(&boot_cpu_data);
556 * partially used pages are not usable - thus
557 * we are rounding upwards:
559 end_pfn = e820_end_of_ram();
560 num_physpages = end_pfn; /* for pfn_valid */
566 init_memory_mapping(0, (end_pfn_map << PAGE_SHIFT));
574 * Initialize the ACPI boot-time table parser (gets the RSDP and SDT).
575 * Call this early for SRAT node setup.
577 acpi_boot_table_init();
580 #ifdef CONFIG_ACPI_NUMA
582 * Parse SRAT to discover nodes.
588 numa_initmem_init(0, end_pfn);
590 contig_initmem_init(0, end_pfn);
593 /* Reserve direct mapping */
594 reserve_bootmem_generic(table_start << PAGE_SHIFT,
595 (table_end - table_start) << PAGE_SHIFT);
598 kernel_end = round_up(__pa_symbol(&_end),PAGE_SIZE);
599 reserve_bootmem_generic(HIGH_MEMORY, kernel_end - HIGH_MEMORY);
602 * reserve physical page 0 - it's a special BIOS page on many boxes,
603 * enabling clean reboots, SMP operation, laptop functions.
605 reserve_bootmem_generic(0, PAGE_SIZE);
607 /* reserve ebda region */
609 reserve_bootmem_generic(ebda_addr, ebda_size);
613 * But first pinch a few for the stack/trampoline stuff
614 * FIXME: Don't need the extra page at 4K, but need to fix
615 * trampoline before removing it. (see the GDT stuff)
617 reserve_bootmem_generic(PAGE_SIZE, PAGE_SIZE);
619 /* Reserve SMP trampoline */
620 reserve_bootmem_generic(SMP_TRAMPOLINE_BASE, PAGE_SIZE);
623 #ifdef CONFIG_ACPI_SLEEP
625 * Reserve low memory region for sleep support.
627 acpi_reserve_bootmem();
629 #ifdef CONFIG_X86_LOCAL_APIC
631 * Find and reserve possible boot-time SMP configuration:
635 #ifdef CONFIG_BLK_DEV_INITRD
636 if (LOADER_TYPE && INITRD_START) {
637 if (INITRD_START + INITRD_SIZE <= (end_pfn << PAGE_SHIFT)) {
638 reserve_bootmem_generic(INITRD_START, INITRD_SIZE);
640 INITRD_START ? INITRD_START + PAGE_OFFSET : 0;
641 initrd_end = initrd_start+INITRD_SIZE;
644 printk(KERN_ERR "initrd extends beyond end of memory "
645 "(0x%08lx > 0x%08lx)\ndisabling initrd\n",
646 (unsigned long)(INITRD_START + INITRD_SIZE),
647 (unsigned long)(end_pfn << PAGE_SHIFT));
653 if (crashk_res.start != crashk_res.end) {
654 reserve_bootmem_generic(crashk_res.start,
655 crashk_res.end - crashk_res.start + 1);
664 * set this early, so we dont allocate cpu0
665 * if MADT list doesnt list BSP first
666 * mpparse.c/MP_processor_info() allocates logical cpu numbers.
668 cpu_set(0, cpu_present_map);
671 * Read APIC and some other early information from ACPI tables.
678 #ifdef CONFIG_X86_LOCAL_APIC
680 * get boot-time SMP configuration:
682 if (smp_found_config)
684 init_apic_mappings();
688 * Request address space for all standard RAM and ROM resources
689 * and also for regions reported as reserved by the e820.
692 e820_reserve_resources();
694 request_resource(&iomem_resource, &video_ram_resource);
698 /* request I/O space for devices used on all i[345]86 PCs */
699 for (i = 0; i < STANDARD_IO_RESOURCES; i++)
700 request_resource(&ioport_resource, &standard_io_resources[i]);
705 #ifdef CONFIG_GART_IOMMU
710 #if defined(CONFIG_VGA_CONSOLE)
711 conswitchp = &vga_con;
712 #elif defined(CONFIG_DUMMY_CONSOLE)
713 conswitchp = &dummy_con;
718 static int __cpuinit get_model_name(struct cpuinfo_x86 *c)
722 if (c->extended_cpuid_level < 0x80000004)
725 v = (unsigned int *) c->x86_model_id;
726 cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
727 cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
728 cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
729 c->x86_model_id[48] = 0;
734 static void __cpuinit display_cacheinfo(struct cpuinfo_x86 *c)
736 unsigned int n, dummy, eax, ebx, ecx, edx;
738 n = c->extended_cpuid_level;
740 if (n >= 0x80000005) {
741 cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
742 printk(KERN_INFO "CPU: L1 I Cache: %dK (%d bytes/line), D cache %dK (%d bytes/line)\n",
743 edx>>24, edx&0xFF, ecx>>24, ecx&0xFF);
744 c->x86_cache_size=(ecx>>24)+(edx>>24);
745 /* On K8 L1 TLB is inclusive, so don't count it */
749 if (n >= 0x80000006) {
750 cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
751 ecx = cpuid_ecx(0x80000006);
752 c->x86_cache_size = ecx >> 16;
753 c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
755 printk(KERN_INFO "CPU: L2 Cache: %dK (%d bytes/line)\n",
756 c->x86_cache_size, ecx & 0xFF);
760 cpuid(0x80000007, &dummy, &dummy, &dummy, &c->x86_power);
761 if (n >= 0x80000008) {
762 cpuid(0x80000008, &eax, &dummy, &dummy, &dummy);
763 c->x86_virt_bits = (eax >> 8) & 0xff;
764 c->x86_phys_bits = eax & 0xff;
769 static int nearby_node(int apicid)
772 for (i = apicid - 1; i >= 0; i--) {
773 int node = apicid_to_node[i];
774 if (node != NUMA_NO_NODE && node_online(node))
777 for (i = apicid + 1; i < MAX_LOCAL_APIC; i++) {
778 int node = apicid_to_node[i];
779 if (node != NUMA_NO_NODE && node_online(node))
782 return first_node(node_online_map); /* Shouldn't happen */
787 * On a AMD dual core setup the lower bits of the APIC id distingush the cores.
788 * Assumes number of cores is a power of two.
790 static void __init amd_detect_cmp(struct cpuinfo_x86 *c)
793 int cpu = smp_processor_id();
797 unsigned apicid = hard_smp_processor_id();
799 unsigned ecx = cpuid_ecx(0x80000008);
801 c->x86_max_cores = (ecx & 0xff) + 1;
803 /* CPU telling us the core id bits shift? */
804 bits = (ecx >> 12) & 0xF;
806 /* Otherwise recompute */
808 while ((1 << bits) < c->x86_max_cores)
812 /* Low order bits define the core id (index of core in socket) */
813 cpu_core_id[cpu] = phys_proc_id[cpu] & ((1 << bits)-1);
814 /* Convert the APIC ID into the socket ID */
815 phys_proc_id[cpu] = phys_pkg_id(bits);
818 node = phys_proc_id[cpu];
819 if (apicid_to_node[apicid] != NUMA_NO_NODE)
820 node = apicid_to_node[apicid];
821 if (!node_online(node)) {
822 /* Two possibilities here:
823 - The CPU is missing memory and no node was created.
824 In that case try picking one from a nearby CPU
825 - The APIC IDs differ from the HyperTransport node IDs
826 which the K8 northbridge parsing fills in.
827 Assume they are all increased by a constant offset,
828 but in the same order as the HT nodeids.
829 If that doesn't result in a usable node fall back to the
830 path for the previous case. */
831 int ht_nodeid = apicid - (phys_proc_id[0] << bits);
832 if (ht_nodeid >= 0 &&
833 apicid_to_node[ht_nodeid] != NUMA_NO_NODE)
834 node = apicid_to_node[ht_nodeid];
835 /* Pick a nearby node */
836 if (!node_online(node))
837 node = nearby_node(apicid);
839 numa_set_node(cpu, node);
841 printk(KERN_INFO "CPU %d/%x(%d) -> Node %d -> Core %d\n",
842 cpu, apicid, c->x86_max_cores, node, cpu_core_id[cpu]);
847 static int __init init_amd(struct cpuinfo_x86 *c)
856 * Disable TLB flush filter by setting HWCR.FFDIS on K8
857 * bit 6 of msr C001_0015
859 * Errata 63 for SH-B3 steppings
860 * Errata 122 for all steppings (F+ have it disabled by default)
863 rdmsrl(MSR_K8_HWCR, value);
865 wrmsrl(MSR_K8_HWCR, value);
869 /* Bit 31 in normal CPUID used for nonstandard 3DNow ID;
870 3DNow is IDd by bit 31 in extended CPUID (1*32+31) anyway */
871 clear_bit(0*32+31, &c->x86_capability);
873 /* On C+ stepping K8 rep microcode works well for copy/memset */
874 level = cpuid_eax(1);
875 if (c->x86 == 15 && ((level >= 0x0f48 && level < 0x0f50) || level >= 0x0f58))
876 set_bit(X86_FEATURE_REP_GOOD, &c->x86_capability);
878 /* Enable workaround for FXSAVE leak */
880 set_bit(X86_FEATURE_FXSAVE_LEAK, &c->x86_capability);
882 r = get_model_name(c);
886 /* Should distinguish Models here, but this is only
887 a fallback anyways. */
888 strcpy(c->x86_model_id, "Hammer");
892 display_cacheinfo(c);
894 /* c->x86_power is 8000_0007 edx. Bit 8 is constant TSC */
895 if (c->x86_power & (1<<8))
896 set_bit(X86_FEATURE_CONSTANT_TSC, &c->x86_capability);
898 /* Multi core CPU? */
899 if (c->extended_cpuid_level >= 0x80000008)
902 /* Fix cpuid4 emulation for more */
903 num_cache_leaves = 3;
908 static void __cpuinit detect_ht(struct cpuinfo_x86 *c)
911 u32 eax, ebx, ecx, edx;
912 int index_msb, core_bits;
913 int cpu = smp_processor_id();
915 cpuid(1, &eax, &ebx, &ecx, &edx);
918 if (!cpu_has(c, X86_FEATURE_HT) || cpu_has(c, X86_FEATURE_CMP_LEGACY))
921 smp_num_siblings = (ebx & 0xff0000) >> 16;
923 if (smp_num_siblings == 1) {
924 printk(KERN_INFO "CPU: Hyper-Threading is disabled\n");
925 } else if (smp_num_siblings > 1 ) {
927 if (smp_num_siblings > NR_CPUS) {
928 printk(KERN_WARNING "CPU: Unsupported number of the siblings %d", smp_num_siblings);
929 smp_num_siblings = 1;
933 index_msb = get_count_order(smp_num_siblings);
934 phys_proc_id[cpu] = phys_pkg_id(index_msb);
936 printk(KERN_INFO "CPU: Physical Processor ID: %d\n",
939 smp_num_siblings = smp_num_siblings / c->x86_max_cores;
941 index_msb = get_count_order(smp_num_siblings) ;
943 core_bits = get_count_order(c->x86_max_cores);
945 cpu_core_id[cpu] = phys_pkg_id(index_msb) &
946 ((1 << core_bits) - 1);
948 if (c->x86_max_cores > 1)
949 printk(KERN_INFO "CPU: Processor Core ID: %d\n",
956 * find out the number of processor cores on the die
958 static int __cpuinit intel_num_cpu_cores(struct cpuinfo_x86 *c)
962 if (c->cpuid_level < 4)
971 return ((eax >> 26) + 1);
976 static void srat_detect_node(void)
980 int cpu = smp_processor_id();
982 /* Don't do the funky fallback heuristics the AMD version employs
984 node = apicid_to_node[hard_smp_processor_id()];
985 if (node == NUMA_NO_NODE)
986 node = first_node(node_online_map);
987 numa_set_node(cpu, node);
990 printk(KERN_INFO "CPU %d -> Node %d\n", cpu, node);
994 static void __cpuinit init_intel(struct cpuinfo_x86 *c)
999 init_intel_cacheinfo(c);
1000 n = c->extended_cpuid_level;
1001 if (n >= 0x80000008) {
1002 unsigned eax = cpuid_eax(0x80000008);
1003 c->x86_virt_bits = (eax >> 8) & 0xff;
1004 c->x86_phys_bits = eax & 0xff;
1005 /* CPUID workaround for Intel 0F34 CPU */
1006 if (c->x86_vendor == X86_VENDOR_INTEL &&
1007 c->x86 == 0xF && c->x86_model == 0x3 &&
1009 c->x86_phys_bits = 36;
1013 c->x86_cache_alignment = c->x86_clflush_size * 2;
1014 if ((c->x86 == 0xf && c->x86_model >= 0x03) ||
1015 (c->x86 == 0x6 && c->x86_model >= 0x0e))
1016 set_bit(X86_FEATURE_CONSTANT_TSC, &c->x86_capability);
1017 set_bit(X86_FEATURE_SYNC_RDTSC, &c->x86_capability);
1018 c->x86_max_cores = intel_num_cpu_cores(c);
1023 static void __cpuinit get_cpu_vendor(struct cpuinfo_x86 *c)
1025 char *v = c->x86_vendor_id;
1027 if (!strcmp(v, "AuthenticAMD"))
1028 c->x86_vendor = X86_VENDOR_AMD;
1029 else if (!strcmp(v, "GenuineIntel"))
1030 c->x86_vendor = X86_VENDOR_INTEL;
1032 c->x86_vendor = X86_VENDOR_UNKNOWN;
1035 struct cpu_model_info {
1038 char *model_names[16];
1041 /* Do some early cpuid on the boot CPU to get some parameter that are
1042 needed before check_bugs. Everything advanced is in identify_cpu
1044 void __cpuinit early_identify_cpu(struct cpuinfo_x86 *c)
1048 c->loops_per_jiffy = loops_per_jiffy;
1049 c->x86_cache_size = -1;
1050 c->x86_vendor = X86_VENDOR_UNKNOWN;
1051 c->x86_model = c->x86_mask = 0; /* So far unknown... */
1052 c->x86_vendor_id[0] = '\0'; /* Unset */
1053 c->x86_model_id[0] = '\0'; /* Unset */
1054 c->x86_clflush_size = 64;
1055 c->x86_cache_alignment = c->x86_clflush_size;
1056 c->x86_max_cores = 1;
1057 c->extended_cpuid_level = 0;
1058 memset(&c->x86_capability, 0, sizeof c->x86_capability);
1060 /* Get vendor name */
1061 cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
1062 (unsigned int *)&c->x86_vendor_id[0],
1063 (unsigned int *)&c->x86_vendor_id[8],
1064 (unsigned int *)&c->x86_vendor_id[4]);
1068 /* Initialize the standard set of capabilities */
1069 /* Note that the vendor-specific code below might override */
1071 /* Intel-defined flags: level 0x00000001 */
1072 if (c->cpuid_level >= 0x00000001) {
1074 cpuid(0x00000001, &tfms, &misc, &c->x86_capability[4],
1075 &c->x86_capability[0]);
1076 c->x86 = (tfms >> 8) & 0xf;
1077 c->x86_model = (tfms >> 4) & 0xf;
1078 c->x86_mask = tfms & 0xf;
1080 c->x86 += (tfms >> 20) & 0xff;
1082 c->x86_model += ((tfms >> 16) & 0xF) << 4;
1083 if (c->x86_capability[0] & (1<<19))
1084 c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
1086 /* Have CPUID level 0 only - unheard of */
1091 phys_proc_id[smp_processor_id()] = (cpuid_ebx(1) >> 24) & 0xff;
1096 * This does the hard work of actually picking apart the CPU stuff...
1098 void __cpuinit identify_cpu(struct cpuinfo_x86 *c)
1103 early_identify_cpu(c);
1105 /* AMD-defined flags: level 0x80000001 */
1106 xlvl = cpuid_eax(0x80000000);
1107 c->extended_cpuid_level = xlvl;
1108 if ((xlvl & 0xffff0000) == 0x80000000) {
1109 if (xlvl >= 0x80000001) {
1110 c->x86_capability[1] = cpuid_edx(0x80000001);
1111 c->x86_capability[6] = cpuid_ecx(0x80000001);
1113 if (xlvl >= 0x80000004)
1114 get_model_name(c); /* Default name */
1117 /* Transmeta-defined flags: level 0x80860001 */
1118 xlvl = cpuid_eax(0x80860000);
1119 if ((xlvl & 0xffff0000) == 0x80860000) {
1120 /* Don't set x86_cpuid_level here for now to not confuse. */
1121 if (xlvl >= 0x80860001)
1122 c->x86_capability[2] = cpuid_edx(0x80860001);
1125 c->apicid = phys_pkg_id(0);
1128 * Vendor-specific initialization. In this section we
1129 * canonicalize the feature flags, meaning if there are
1130 * features a certain CPU supports which CPUID doesn't
1131 * tell us, CPUID claiming incorrect flags, or other bugs,
1132 * we handle them here.
1134 * At the end of this section, c->x86_capability better
1135 * indicate the features this CPU genuinely supports!
1137 switch (c->x86_vendor) {
1138 case X86_VENDOR_AMD:
1142 case X86_VENDOR_INTEL:
1146 case X86_VENDOR_UNKNOWN:
1148 display_cacheinfo(c);
1152 select_idle_routine(c);
1156 * On SMP, boot_cpu_data holds the common feature set between
1157 * all CPUs; so make sure that we indicate which features are
1158 * common between the CPUs. The first time this routine gets
1159 * executed, c == &boot_cpu_data.
1161 if (c != &boot_cpu_data) {
1162 /* AND the already accumulated flags with these */
1163 for (i = 0 ; i < NCAPINTS ; i++)
1164 boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
1167 #ifdef CONFIG_X86_MCE
1170 if (c == &boot_cpu_data)
1175 numa_add_cpu(smp_processor_id());
1180 void __cpuinit print_cpu_info(struct cpuinfo_x86 *c)
1182 if (c->x86_model_id[0])
1183 printk("%s", c->x86_model_id);
1185 if (c->x86_mask || c->cpuid_level >= 0)
1186 printk(" stepping %02x\n", c->x86_mask);
1192 * Get CPU information for use by the procfs.
1195 static int show_cpuinfo(struct seq_file *m, void *v)
1197 struct cpuinfo_x86 *c = v;
1200 * These flag bits must match the definitions in <asm/cpufeature.h>.
1201 * NULL means this bit is undefined or reserved; either way it doesn't
1202 * have meaning as far as Linux is concerned. Note that it's important
1203 * to realize there is a difference between this table and CPUID -- if
1204 * applications want to get the raw CPUID data, they should access
1205 * /dev/cpu/<cpu_nr>/cpuid instead.
1207 static char *x86_cap_flags[] = {
1209 "fpu", "vme", "de", "pse", "tsc", "msr", "pae", "mce",
1210 "cx8", "apic", NULL, "sep", "mtrr", "pge", "mca", "cmov",
1211 "pat", "pse36", "pn", "clflush", NULL, "dts", "acpi", "mmx",
1212 "fxsr", "sse", "sse2", "ss", "ht", "tm", "ia64", NULL,
1215 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1216 NULL, NULL, NULL, "syscall", NULL, NULL, NULL, NULL,
1217 NULL, NULL, NULL, NULL, "nx", NULL, "mmxext", NULL,
1218 NULL, "fxsr_opt", "rdtscp", NULL, NULL, "lm", "3dnowext", "3dnow",
1220 /* Transmeta-defined */
1221 "recovery", "longrun", NULL, "lrti", NULL, NULL, NULL, NULL,
1222 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1223 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1224 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1226 /* Other (Linux-defined) */
1227 "cxmmx", NULL, "cyrix_arr", "centaur_mcr", NULL,
1228 "constant_tsc", NULL, NULL,
1229 "up", NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1230 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1231 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1233 /* Intel-defined (#2) */
1234 "pni", NULL, NULL, "monitor", "ds_cpl", "vmx", "smx", "est",
1235 "tm2", NULL, "cid", NULL, NULL, "cx16", "xtpr", NULL,
1236 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1237 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1239 /* VIA/Cyrix/Centaur-defined */
1240 NULL, NULL, "rng", "rng_en", NULL, NULL, "ace", "ace_en",
1241 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1242 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1243 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1245 /* AMD-defined (#2) */
1246 "lahf_lm", "cmp_legacy", "svm", NULL, "cr8_legacy", NULL, NULL, NULL,
1247 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1248 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1249 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1251 static char *x86_power_flags[] = {
1252 "ts", /* temperature sensor */
1253 "fid", /* frequency id control */
1254 "vid", /* voltage id control */
1255 "ttp", /* thermal trip */
1259 /* nothing */ /* constant_tsc - moved to flags */
1264 if (!cpu_online(c-cpu_data))
1268 seq_printf(m,"processor\t: %u\n"
1270 "cpu family\t: %d\n"
1272 "model name\t: %s\n",
1273 (unsigned)(c-cpu_data),
1274 c->x86_vendor_id[0] ? c->x86_vendor_id : "unknown",
1277 c->x86_model_id[0] ? c->x86_model_id : "unknown");
1279 if (c->x86_mask || c->cpuid_level >= 0)
1280 seq_printf(m, "stepping\t: %d\n", c->x86_mask);
1282 seq_printf(m, "stepping\t: unknown\n");
1284 if (cpu_has(c,X86_FEATURE_TSC)) {
1285 unsigned int freq = cpufreq_quick_get((unsigned)(c-cpu_data));
1288 seq_printf(m, "cpu MHz\t\t: %u.%03u\n",
1289 freq / 1000, (freq % 1000));
1293 if (c->x86_cache_size >= 0)
1294 seq_printf(m, "cache size\t: %d KB\n", c->x86_cache_size);
1297 if (smp_num_siblings * c->x86_max_cores > 1) {
1298 int cpu = c - cpu_data;
1299 seq_printf(m, "physical id\t: %d\n", phys_proc_id[cpu]);
1300 seq_printf(m, "siblings\t: %d\n", cpus_weight(cpu_core_map[cpu]));
1301 seq_printf(m, "core id\t\t: %d\n", cpu_core_id[cpu]);
1302 seq_printf(m, "cpu cores\t: %d\n", c->booted_cores);
1308 "fpu_exception\t: yes\n"
1309 "cpuid level\t: %d\n"
1316 for ( i = 0 ; i < 32*NCAPINTS ; i++ )
1317 if (cpu_has(c, i) && x86_cap_flags[i] != NULL)
1318 seq_printf(m, " %s", x86_cap_flags[i]);
1321 seq_printf(m, "\nbogomips\t: %lu.%02lu\n",
1322 c->loops_per_jiffy/(500000/HZ),
1323 (c->loops_per_jiffy/(5000/HZ)) % 100);
1325 if (c->x86_tlbsize > 0)
1326 seq_printf(m, "TLB size\t: %d 4K pages\n", c->x86_tlbsize);
1327 seq_printf(m, "clflush size\t: %d\n", c->x86_clflush_size);
1328 seq_printf(m, "cache_alignment\t: %d\n", c->x86_cache_alignment);
1330 seq_printf(m, "address sizes\t: %u bits physical, %u bits virtual\n",
1331 c->x86_phys_bits, c->x86_virt_bits);
1333 seq_printf(m, "power management:");
1336 for (i = 0; i < 32; i++)
1337 if (c->x86_power & (1 << i)) {
1338 if (i < ARRAY_SIZE(x86_power_flags) &&
1340 seq_printf(m, "%s%s",
1341 x86_power_flags[i][0]?" ":"",
1342 x86_power_flags[i]);
1344 seq_printf(m, " [%d]", i);
1348 seq_printf(m, "\n\n");
1353 static void *c_start(struct seq_file *m, loff_t *pos)
1355 return *pos < NR_CPUS ? cpu_data + *pos : NULL;
1358 static void *c_next(struct seq_file *m, void *v, loff_t *pos)
1361 return c_start(m, pos);
1364 static void c_stop(struct seq_file *m, void *v)
1368 struct seq_operations cpuinfo_op = {
1372 .show = show_cpuinfo,
1375 #ifdef CONFIG_INPUT_PCSPKR
1376 #include <linux/platform_device.h>
1377 static __init int add_pcspkr(void)
1379 struct platform_device *pd;
1382 pd = platform_device_alloc("pcspkr", -1);
1386 ret = platform_device_add(pd);
1388 platform_device_put(pd);
1392 device_initcall(add_pcspkr);